← AI Hashrate

RX 7600 8GB

AMD · 8 GB GDDR6 · 288 GB/s · 165 W · MSRP $269 · Search Amazon (affiliate)

The RX 7600 8GB is a AMD accelerator with 8 GB GDDR6 and about 288 GB/s peak memory bandwidth. AI Hashrate estimates local LLM decode (batch 1) at 8K context: fit counts weights + KV cache + 1 GB runtime overhead — not weights alone. At Q4, about 17 curated models fit fully on this card; at FP16, about 4 fit. Top Q4 speeds: Qwen3-0.6B ≈ 229.1 tok/s (estimated, fits); gemma-3-1b-it ≈ 183.3 tok/s (estimated, fits); Llama-3.2-1B-Instruct ≈ 152.7 tok/s (estimated, fits); Qwen3-1.7B ≈ 91.6 tok/s (estimated, fits); Llama-3.2-3B-Instruct ≈ 57.3 tok/s (estimated, fits). Relative ranking is more reliable than absolute tok/s. See methodology for the bandwidth formula and measured-anchor policy. Methodology.

Table: decode speed estimates (Q4 / FP16). Measured rows override estimates. MSRP is list price, not live retail. Context default 8K.

ModelParams (B)Quanttok/sFits?
Qwen3-0.6B0.8Q4229.1 est.Yes
gemma-3-1b-it1.0Q4183.3 est.Yes
Llama-3.2-1B-Instruct1.2Q4152.7 est.Yes
Qwen3-1.7B2.0Q491.6 est.Yes
Qwen3-0.6B0.8FP1663.0 est.Yes
Llama-3.2-3B-Instruct3.2Q457.3 est.Yes
gpt-oss-20b21.5Q450.7 est.No
gemma-3-1b-it1.0FP1650.4 est.Yes
Phi-4-mini-instruct3.8Q448.2 est.Yes
Phi-4-mini-reasoning3.8Q448.2 est.Yes
gemma-3-4b-it4.3Q447.0 est.Yes
Qwen3-4B-Instruct-25074.0Q445.8 est.Yes
Qwen3-4B-Thinking-25074.0Q445.8 est.Yes
Llama-3.2-3B-Instruct3.2FP1644.6 est.No
Llama-3.2-1B-Instruct1.2FP1642.0 est.Yes
Qwen3-30B-A3B30.5Q433.1 est.No
Qwen3-Coder-30B-A3B-Instruct30.5Q433.1 est.No
Qwen3-30B-A3B-Instruct-250730.5Q433.1 est.No
Phi-4-mini-instruct3.8FP1627.7 est.No
Phi-4-mini-reasoning3.8FP1627.7 est.No
gemma-3-12b-it12.2Q427.3 est.No
Mistral-7B-Instruct-v0.37.2Q425.5 est.Yes
Qwen3-1.7B2.0FP1625.2 est.Yes
Qwen2.5-7B-Instruct7.6Q424.1 est.Yes
Qwen2.5-Coder-7B-Instruct7.6Q424.1 est.Yes
DeepSeek-R1-Distill-Qwen-7B7.6Q424.1 est.Yes
Qwen3-4B-Instruct-25074.0FP1624.0 est.No
Qwen3-4B-Thinking-25074.0FP1624.0 est.No
Llama-3.1-8B-Instruct8.0Q422.9 est.Yes
Qwen3-8B8.2Q422.4 est.Yes
DeepSeek-R1-0528-Qwen3-8B8.2Q422.4 est.Yes
gemma-3-4b-it4.3FP1621.9 est.No
Phi-414.7Q415.8 est.No
Phi-4-reasoning14.7Q415.8 est.No
Qwen3-14B14.8Q415.5 est.No
Qwen2.5-14B-Instruct14.8Q415.5 est.No
Qwen2.5-Coder-14B-Instruct14.8Q415.5 est.No
DeepSeek-R1-Distill-Qwen-14B14.8Q415.5 est.No
GLM-4.7-Flash31.2Q46.6 est.No
Mistral-7B-Instruct-v0.37.2FP164.5 est.No
Devstral-Small-250723.6Q44.1 est.No
Qwen2.5-7B-Instruct7.6FP163.9 est.No
Qwen2.5-Coder-7B-Instruct7.6FP163.9 est.No
DeepSeek-R1-Distill-Qwen-7B7.6FP163.9 est.No
Mistral-Small-3.2-24B-Instruct-250624.0Q43.9 est.No
Magistral-Small-250924.0Q43.9 est.No
Llama-3.1-8B-Instruct8.0FP163.3 est.No
Qwen3-8B8.2FP163.1 est.No
DeepSeek-R1-0528-Qwen3-8B8.2FP163.1 est.No
Mixtral-8x7B-Instruct-v0.146.7Q43.0 est.No
gemma-3-27b-it27.4Q42.7 est.No
Qwen3-32B32.8Q41.6 est.No
Qwen2.5-32B-Instruct32.8Q41.6 est.No
Qwen2.5-Coder-32B-Instruct32.8Q41.6 est.No
DeepSeek-R1-Distill-Qwen-32B32.8Q41.6 est.No
gemma-3-12b-it12.2FP161.0 est.No